What the evidence does support is narrower: losses dominate several large studies of frequent day traders and retail leveraged products. In Taiwan, more than eight in ten day traders lost money in a typical six-month period; a later 15-year analysis found fewer than 1% predictably earned positive abnormal returns after fees. In one Brazilian equity-futures cohort, 97% of people who persisted beyond 300 trading days lost money. These results are serious, but none is a global census of every trader in 2026.
| Question this guide can answer | What each cited source measured and what its result means |
| Question it cannot answer | Your personal probability of profit or a universal global percentage |
| Previous TSB 8,400-user claim | Not verified — no frozen extract and reproducible calculation report located in the project evidence |
| Best personal benchmark | Net expectancy, uncertainty range, drawdown and rule adherence from your own stable sample |
| Evidence date | Sources rechecked September 7, 2026; original study periods remain historical |
A useful statistic needs a population, product, market, inclusion rule, observation window, cost treatment and outcome definition. “Profitable in one quarter,” “positive abnormal return next year” and “earned enough to live on after 300 days” are different endpoints. This guide keeps them separate.
What Percentage of Traders Are Actually Profitable?
The honest global answer is Not verified. No source reviewed here observes every self-directed stock, futures, options, forex, CFD and crypto trader worldwide under one definition through 2026. The strongest available evidence is a set of bounded estimates. Quote the bound with the number.
What the primary sources actually report
| Source | Population and period | Published result | Valid interpretation |
|---|---|---|---|
| Barber, Lee, Liu & Odean | Taiwan Stock Exchange day traders, 1995–1999 | More than 8 in 10 lost money in a typical six-month period | Net day-trading outcome in that market and horizon; not a worldwide lifetime rate |
| Barber et al., 2014 | All Taiwan Stock Exchange day traders, 1992–2006 | About 20% positive after fees in an average year; fewer than 1% predictably profitable next year | One-year realized profit and persistent skill are different statistics |
| Chague, De-Losso & Giovannetti | New Brazilian equity-futures day traders who began in 2013–2015; persistent subgroup >300 days | 97% of the persistent subgroup lost; 1.1% earned more than Brazil’s minimum wage | A severe result for persistent day trading in one futures market, not every retail strategy |
| ESMA / national regulators | Retail CFD accounts across several EU jurisdictions, analyses available by 2018 | 74%–89% of accounts typically lost money | Account-level leveraged-CFD evidence; not a trader-level global day-trading rate |
| CFTC advisory | Registered US OTC forex dealer disclosures, Q2 2021–Q1 2022 | Roughly one-third of accounts profitable and two-thirds not profitable | Quarterly account results after specified charges; percentages vary by dealer and quarter |
The Taiwan 2014 paper is the cleanest demonstration of why one headline can mislead. Around 20% of active day traders earned positive abnormal returns after fees in an average year, yet fewer than 1% showed performance that reliably predicted positive net abnormal returns in the following year. A profitable year can reflect skill, luck or both; persistence asks a harder question.
Why the numbers diverge
- Population: every account, active traders, new entrants and traders who persisted 300 days select very different people.
- Horizon: a profitable quarter is easier to observe by chance than repeated profit across years.
- Outcome: positive P&L, market-beating abnormal return and income above a wage threshold are not synonyms.
- Costs: commissions, spread, tax, financing, slippage and data fees can turn gross profit into net loss.
- Product: unlevered stock investing, intraday futures, OTC forex and CFDs have different mechanics and risks.
How Much Do Day Traders Make? “Average Salary” Is the Wrong Denominator
A self-directed day trader normally has trading profit or loss, not a salary. Salary websites often mix proprietary-firm employees, brokerage roles and self-employed traders, then publish a number with no verified population. That cannot answer what an independent retail trader earns.
The Brazilian study offers a bounded income comparison rather than a universal salary. Among people in its persistent subgroup, 1.1% earned more than the Brazilian minimum wage and 0.5% earned more than the initial salary of a bank teller, with substantial risk. Those thresholds belong to Brazil and the study period; converting them into a 2026 US “average salary” would be false precision.
A 2% monthly return is useful here only as arithmetic, not as a normal, disciplined, or promised outcome: 2% of $5,000 is $100, while 2% of $100,000 is $2,000 before costs and taxes. The percentage is identical; the capital, drawdown exposure, cash result, and uncertainty are not. Never annualize the example or use it to infer that either account will produce the return.
For an individual record, report net trading P&L after all direct costs, the capital and margin required, hours spent, maximum drawdown and an uncertainty range. A $20,000 gain on $500,000 with large drawdowns is not economically comparable to the same gain on $50,000 with lower risk. Do not annualize a short winning streak as salary.
There Is No Credible Universal Average Trader Win Rate
None of the population studies above establishes a global average percentage of winning trades. They generally evaluate account returns or trader profitability. A strategy can win often and still lose money if losses are larger than wins; another can win less often and remain profitable when winners are sufficiently larger.
Use expectancy, not win rate alone
E = (p × average win) − ((1 − p) × average loss) − average cost
Here, p is the observed win rate, win and loss are positive magnitudes, and cost includes commissions, spread, fees and an honest slippage estimate. In risk units, a 40% win rate with a 2R average win and 1R average loss has 0.20R gross expectancy: (0.40 × 2) − (0.60 × 1). Costs must still be subtracted.
| Average win : average loss | Gross breakeven win rate | With 0.05R average cost |
|---|---|---|
| 0.5 : 1 | 66.7% | 70.0% |
| 1 : 1 | 50.0% | 52.5% |
| 1.5 : 1 | 40.0% | 42.0% |
| 2 : 1 | 33.3% | 35.0% |
| 3 : 1 | 25.0% | 26.3% |
The cost-adjusted column assumes every win and loss lands exactly at its stated size. Real distributions contain partial exits, gaps and outliers, so calculate from trade-level net P&L rather than substituting planned targets.
A win rate needs an uncertainty interval
Fifty winners in 100 independent trades produces a 50% observed win rate, but an approximate 95% Wilson interval is about 40%–60%. Five hundred wins in 1,000 trades narrows that interval to roughly 47%–53%. Dependence between trades, regime changes and strategy edits make the effective sample smaller than the raw count.
A 77% hit rate for two weeks can be a truthful description of one short record and still be useless as proof of durable skill. Report the number of eligible trades, market, setup version, costs, exclusions and confidence interval; then test the same definition on later data. Two weeks is a time label, not a denominator.
An 8% profit over 30 days can be equally real for one account and equally weak as a durable benchmark. Keep the opening capital, net-dollar result, eligible trades, costs, drawdown, deposits or withdrawals, leverage and exact cutoff beside the percentage. A 30-day observation is evidence about that defined window—not an annual return, salary, population average or promise.
That is why a web benchmark such as “average win rate 47%” is not useful without a source and compatible strategy. Track the estimate by setup, market, session and stable strategy version, and show the sample size beside it.
Reward-to-Risk Ratios Matter, but Planned R:R Is Not Realized R:R
A chart setup may target 2R, yet early exits can produce a 1.2R average winner while slippage pushes the average loss beyond 1R. The realized distribution—not the label written before entry—belongs in expectancy.
| Metric | Definition | Main failure mode |
|---|---|---|
| Win rate | Winning closed trades ÷ classified closed trades | Breakevens and partial exits classified inconsistently |
| Average win / loss | Mean positive result / absolute mean negative result | One large outlier dominates a small sample |
| Profit factor | Gross profit ÷ absolute gross loss | Undefined with no losses; unstable in short samples |
| Expectancy | Average net outcome per trade or risk unit | Costs omitted or changing position size mixed in dollars |
| Maximum drawdown | Largest peak-to-trough decline in the equity sequence | Trade order and open equity ignored |
A positive historical expectancy is not proof of a future edge. Recompute it out of sample, keep the strategy definition stable, and report drawdown beside the mean. Two strategies can share expectancy and have radically different loss streaks.
Does Journaling Actually Help Performance?
The absence of measurement leaves a trader unable to distinguish an edge from a story. Journaling fixes that evidence gap: it creates the record needed to calculate net expectancy, identify rule breaches, segment setups and audit whether actual execution matches the plan. That practical value does not establish a causal profitability uplift.
The previous version claimed 38% profitability for daily journalers, 19% for rare journalers and an 18% six-month profit-factor improvement. No frozen anonymized extract, cohort definition, deduplication rule, cost policy, analysis code or reproducible report supporting those figures was located in the project evidence for this revision. They are therefore removed and classified as Not verified.
Correlation is not causation
Even a real association would face self-selection: disciplined or already-profitable traders may be more likely to keep complete journals. Survivorship, paid-product usage, missing losing accounts and strategy changes can amplify the difference. A credible causal claim would need a defined intervention, comparable control group, prespecified outcome, attrition accounting and sufficient follow-up.
The supported conclusion is narrower: a journal makes your own process observable. Whether it improves outcomes depends on what you record, whether reviews produce a testable change, and whether you evaluate that change on later data. See how to analyze trading performance for the calculation workflow and the journal comparison for product features.
Performance by Asset Class Cannot Be Ranked from These Studies
The cited evidence covers different populations rather than a controlled contest between markets. Taiwan equity day traders, Brazilian equity-futures day traders, EU CFD accounts and US OTC forex accounts differ in leverage, regulation, costs, session structure, access and observation dates. Their percentages cannot be placed in a league table to conclude that stocks, forex, futures or crypto are “more profitable.”
The previous asset-class table—35% profitable stocks/options, 33% forex, 29% futures and 27% crypto—depended on the same unverified internal dataset and is removed. For a useful personal comparison, hold the trader and metric definition constant, net all costs, require adequate samples in each market and report uncertainty. Otherwise the table mostly measures who chose each market.
The Hidden Deal-Breaker: Survivorship and Selection Bias
A dataset requiring 90 days of activity necessarily excludes anyone who stopped before day 90. A journal-app sample excludes non-users and may miss trades users decline to import. A “persistent trader” study deliberately answers a question about people who remained active, not all starters. None of these designs is automatically wrong; the error is changing the denominator after reading the result.
How bias changes the headline
- Attrition: closed and abandoned accounts may disappear from a current-user snapshot.
- Minimum-activity filters: requiring trades or days can remove both early failures and cautious inactive users.
- Missing data: manual logs may omit fees, partial fills or embarrassing losses.
- Multiple accounts: an account-level denominator can count one trader more than once.
- Strategy drift: combining changing systems creates a statistic for no actual strategy.
The remedy is not to invent a “bias-adjusted honest number.” Preserve the source’s denominator, state exclusions, and resist generalizing beyond them.
Three Mistakes Traders Make with Performance Statistics
Mistake 1: Turning a population result into a personal probability
“97% lost in this persistent Brazilian cohort” does not mean every reader has an immutable 3% chance. It is evidence about a defined market, entry cohort, rule set and period. Your estimate requires your own process and data—but the severe base-rate evidence is a reason to demand stronger proof before risking money, not a reason to ignore the study.
Mistake 2: Comparing win rate without payoff and costs
A 60% win rate can lose when average losses exceed average wins. A 35% win rate can profit with sufficiently large winners. Compare net expectancy, distribution, drawdown and sample uncertainty together. Do not repair an unfavorable realized distribution by quoting planned R:R.
Mistake 3: Treating a short backtest or winning month as income
Optimization, overlapping trades and market regime can make a backtest look more certain than it is. Keep a holdout period, include realistic costs and avoid annualizing the best month. If dozens of variants were tried, the selected result needs a stronger out-of-sample test because selection itself creates false winners.
Who Should Skip Aggregate Benchmarks and Build a Personal Baseline
Aggregate percentages are especially weak for a low-frequency swing strategy, a market absent from the source, a recently changed system, or an account constrained by prop-firm rules. In those cases, use the external evidence as a risk prior and your own stable sample for operating decisions.
| Report field | Required definition |
|---|---|
| Observation window | Exact start/end dates and market regimes represented |
| Sample | Closed trades, open positions, exclusions and strategy version |
| Net result | Realized and unrealized P&L with commissions, spread, financing and fees |
| Expectancy | Per trade and per unit of initial risk, plus uncertainty or resampling range |
| Drawdown | Peak-to-trough equity including open-equity path when relevant |
| Robustness | Out-of-sample period, sensitivity to costs and largest-trade dependence |
Prop-firm candidates should additionally model the exact daily, maximum-loss, consistency and payout rules. The prop-firm pass-rate guide treats that as a different population. No firm or program is part of the decision on this educational page, so a catalog card would add false specificity and is intentionally absent.
Methodology, Source Map and Limitations
This revision uses primary academic papers and official regulator material. Each number is stored with its population, market, sample period, unit of analysis, horizon, cost treatment and endpoint. We do not average the percentages because they estimate different things.
TSB scale and analysis cohorts are different layers. Trader's Second Brain has processed 600K+ imported trades cumulatively. That is meaningful ingestion scale, not an automatic denominator for every result on this page. A first-party statistic must separately show current/queryable rows, named filters, eligible n, excluded rows, unit of analysis, and the denominator behind each percentage. For example, 3,000 eligible shorts plus 2,000 eligible longs describes a 5,000-trade direction-classified cohort—not all cumulative imports.
The study data are historical. “2026” describes the publication and review year, not the sample year and not a claim that market outcomes are unchanged. The CFTC advisory’s roughly one-third profitable figure is based on Q2 2021 through Q1 2022 dealer disclosures; ESMA’s 74%–89% range summarizes regulatory analyses available when its 2018 intervention was adopted. Current provider warnings can move each quarter.
The prior TSB figures are excluded until a versioned, privacy-safe evidence package exists with a query or code path, frozen aggregate output, definitions, exclusion and deduplication logic, cost coverage, missing-data analysis and a reviewed release date. Reintroducing them requires a fresh editorial review; a marketing claim or database count alone is insufficient.
For the broader process, read the editorial methodology. For a practical metric workflow, use the expectancy formula and maximum drawdown guide.
Final Verdict: The Evidence Is Severe, but the Denominators Matter
Large primary sources do not support a reassuring view of frequent retail trading. More than eight in ten Taiwan day traders lost over a typical six-month period in one study; fewer than 1% of the broader Taiwan day-trader population showed predictable net abnormal profit in a later 15-year analysis; and 97% of a persistent Brazilian equity-futures cohort lost money. Regulators have also documented majority losses in retail CFDs and OTC forex accounts.
Those findings do not combine into one global 2026 percentage, and they do not supply an average salary, universal win rate or personal forecast. The most useful conclusion is methodological: keep the population attached to every number, subtract all costs, separate a lucky period from persistent skill, and judge your own strategy by net expectancy, uncertainty and drawdown.
Journaling is valuable because it makes that test possible. A journal is not evidence that its users are profitable, and no causal uplift is claimed here. If your data cannot survive a frozen calculation and an out-of-sample check, the correct result is still Not verified.
Build a benchmark from your own trades
Import a stable sample, include costs, and review expectancy and drawdown without relying on a universal success-rate slogan.
Analyze your trades →